JP2001345299A - Apparatus and method for measuring end point of process and polishing apparatus and method for producing semiconductor device and medium recording signal processing program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a measuring apparatus for detecting the end point of a process for removing a film on a wafer in a process for producing a semiconductor device, e.g. an LSI, in situ with high accuracy even when a pattern is present on the surface, a polishing layer is not varied clearly, or disturbance due to slurry or difference of measuring position is present. SOLUTION: Two or more feature quantities are extracted from a signal waveform obtained by irradiating the surface of a substrate with white light and detecting one or both of reflection signal light or transmission signal light thereof. When the end point of a process is measured by performing logical operation using two or more feature quantities, tuning is performed additionally using a fuzzy rule.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a process for manufacturing a semiconductor device such as an LSI, for example, and detects an end point of the process in a process of forming a film on a semiconductor wafer or a process of polishing a thin film on the wafer. The present invention relates to a measuring apparatus, a measuring method, a polishing apparatus, a semiconductor device manufacturing method, and a recording medium on which a measuring method program is recorded.

[0002]

2. Description of the Related Art High density semiconductor devices have continued to evolve without showing any limitations.
Methods are being developed. One of them is a multilayer wiring, and the technical problems associated with this include flattening a global (relatively large area) device surface on a semiconductor wafer and wiring between upper and lower layers.

[0003] Shortening the exposure wavelength in the lithography process,
Furthermore, high NA (numerical aperture)
Considering the reduction in the depth of focus during exposure accompanying the above, there is a great demand for the accuracy of flattening the interlayer layer at least in the range of the exposure area. There is also a great demand for so-called inlays (plugs, damascenes), which are burying of metal electrode layers in order to realize multilayer wiring, and in this case, it is necessary to remove and flatten an extra metal layer after lamination of the metal layers. . A polishing process called CMP is drawing attention as an efficient planarization technique for these large areas. CMP (Ch
emical Mechanical Polishing or Planarization)
Is a process that removes the surface layer of a wafer by using both physical polishing and chemical polishing (dissolution with a polishing agent solution), and is the leading global flattening and electrode forming technology. Has become a candidate. Specifically, using an abrasive called a slurry in which abrasive particles (typically silica, alumina, cerium oxide, etc.) are dispersed in a soluble solvent of the object to be polished, such as an acid or an alkali, and using an appropriate polishing pad Then, the polishing is advanced by pressing the wafer surface and rubbing by relative motion. By making the pressure and the relative movement speed uniform over the entire surface of the wafer, uniform polishing within the surface becomes possible.

FIG. 12 is a schematic view of a conventional CMP polishing apparatus. Is wafer 2 mounted on the polishing head 1 is pressed against the polishing pad 3 while rotating at an angular velocity omega _H. Plate 4 where the polishing pad is fixed, rotates at an angular velocity omega _T.
An abrasive supply mechanism 16 is provided between the wafer 2 and the polishing pad 3.
(Slurry) 17 is supplied from the
The polished surface of the wafer 2 is polished by the chemical action and the physical action of the polishing pad 3. Polishing rate v of any point of the wafer 2 surface within the distance from the center of the platen 4 to the center of the polishing head 1 r _C, the distance from the center of the polishing head 1 to the polishing point When r _H v = r _C・ ω _T −r _H・ (ω
_H −ω _T ), the polishing speed is constant regardless of the position in the wafer 2 when ω _H = ω _T.

[0005] One of the major requirements of this process is as follows.
There is detection of the end point of the polishing process. In particular, the detection of the polishing end point (in-situ) during the polishing process is greatly demanded in order to improve the process efficiency.

As this detection method, a general film thickness measuring device is often used for detecting the end point of the polishing process. A minute blank portion (where no device pattern is present) of the wafer cleaned after the process is selected as a measurement location, and detection and measurement are performed by various methods.

In the polishing and flattening step, as a faster monitoring method, a method of detecting a fluctuation in friction when polishing proceeds to a layer different from the target polishing layer by a change in motor torque of wafer rotation or pad rotation. is there.

Another method is to irradiate a laser beam to a wafer surface, track the time variation of reflected light intensity using optical interference, and measure the film thickness. In many cases, the end point is determined when the intensity changes over time and reaches a predetermined value.However, due to the uncertainty depending on the device pattern of the wafer, errors due to the measurement position, and the effects of signal noise, the process ends. It is pointed out that it is difficult to judge points clearly.

[0009]

There are various methods for detecting the end point in the above-described CMP process, but no definitive method has yet been found.

For example, measurement with a film thickness measuring instrument can obtain sufficient accuracy and reliable data, but the apparatus itself becomes large-scale, measurement takes time, and feedback to the process is required. Become slow.

Although the method of detecting the process end point by the motor torque is simple and high-speed, it is effective only when detecting that the layer has clearly changed to a different type as the process end point, and is more accurate. Is not enough.

Furthermore, the method of irradiating a laser beam onto a wafer surface combines the uncertainty and error of the measurement position depending on the device pattern type of the wafer, and the influence of signal noise caused by slurry and the like. It has been pointed out that it is difficult to clearly determine the process end point due to signal disturbance.

[0013] The present invention solves the above-mentioned problems, and enables the in-situ measurement of the polishing end point even if the signal is disturbed and the polishing layer does not change to a clearly different type. Provided are a measuring device, a measuring method, a polishing device, and a recording medium on which a semiconductor device manufacturing method and a measuring method are recorded.

[0014]

Means for Solving the Problems In order to solve the above problems, the present invention firstly sets a step of forming an insulating film or a metal electrode film on a substrate, or a step end point in a step of removing the film. A measurement device that irradiates the substrate surface with light and measures a signal waveform obtained by detecting one or both of the reflected signal light and the transmitted signal light, and extracts two or more feature amounts from the signal waveform. There is provided a measuring apparatus, comprising: a feature amount extraction unit; and a logic operation unit that performs a logic operation using the two or more feature amounts to determine a process end point.

The measurement performed by the measuring device in this means is mainly detection of the process end point. Secondly, the signal waveform is a spectral waveform, and the feature amount is a maximum value, a maximum value, a minimum value, a minimum value, a maximum value / a minimum value in the signal waveform, Maximum value / minimum value, and | maximum value−minimum value |
And each | maximum value− for a plurality of the maximum value / minimum value pairs.
A feature value comprising an addition value of the minimum value |, an integrated value of the signal waveform, a group of one and two time differential coefficients of each of the feature amounts, and a group of positive and negative signs of the time differential coefficient 2. The measuring device according to claim 1, wherein the measuring device is selected from the group.

Thirdly, there is provided the measuring device according to any one of claims 1 to 2, wherein the logical operation unit makes the determination using fuzzy inference. Fourthly, there is provided a measuring apparatus according to claim 3, wherein a membership function used in the fuzzy inference is tuned during measurement by a value calculated from the feature amount.

Fifth, the step of forming an insulating film or a metal electrode film on the substrate, or the end point of the step of removing the film, is performed by irradiating the surface of the substrate with light and reflecting the reflected signal light or the transmitted signal light. It is a measuring device that measures from a change in a feature value extracted from a signal waveform obtained by detecting one or both of the above, comprising a feature value extraction unit that extracts a feature value from the signal waveform, and the signal waveform, Is a spectral waveform, and the feature amount is | maximum value-minimum value || for a pair of adjacent maximum value / minimum value in the signal waveform, or | maximum value |-each of a plurality of the maximum value / minimum value pairs. A measuring device characterized by being an added value of the minimum value | or an integrated value of the signal waveform.

Sixth, the characteristic amount is extracted from a waveform obtained by normalizing the signal waveform.
A measurement device according to any one of the preceding claims. Seventhly, there is provided the measuring device according to any one of claims 1 to 6, wherein the characteristic amount is extracted from a waveform obtained by rotating and correcting the signal waveform.

Eighth, the step of forming an insulating film or a metal electrode film on the substrate or the step of removing the film is determined by irradiating the substrate surface with light and reflecting or transmitting signal light. A method of measuring from a signal waveform obtained by detecting one or both of the above, and extracting two or more features from the signal waveform, and performing a logical operation using the two or more features to determine And a measuring method comprising the steps of:

Ninth, a step of forming an insulating film or a metal electrode film on a substrate or an end point of the step of removing the film is performed by irradiating the substrate surface with light and reflecting or transmitting the reflected signal light or the transmitted signal light. Is a measurement method for measuring by a change in a feature value extracted from a signal waveform obtained by detecting one or both of the signal waveforms, wherein the signal waveform is a spectral waveform, and the feature value is included in the signal waveform. | Maximum value-minimum value | for an adjacent local maximum value / minimum value pair, or the sum of each | maximum value-minimum value | of a plurality of the maximum value / minimum value pairs, or the integral value of the signal waveform. And a measuring method characterized by the following.

Here, the maximum value / minimum value and the maximum maximum value / minimum minimum value in the second, fifth, and ninth means are respectively a value obtained by dividing the maximum value by the minimum value, and a maximum value. Is divided by the minimum minimum value, and | maximum value−minimum value | indicates the absolute value of the value obtained by subtracting the minimum value from the maximum value.

Tenthly, it comprises a holding portion for holding a substrate, a polishing body, and the measuring device according to any one of claims 1 to 7, wherein an abrasive is provided between the substrate and the polishing body. When a load is applied between the substrate and the polishing body in a state where the substrate is interposed and a substrate is polished by applying a relative motion between the two, the end point of the process can be measured. The present invention provides a polishing apparatus characterized by the following.

Eleventhly, there is provided a method of manufacturing a semiconductor device, comprising a step of polishing a surface of a semiconductor wafer by using the polishing apparatus according to the tenth aspect. Twelfth, a machine readable recording of a signal processing program for causing a computer to function as the “feature amount extraction unit and logical operation unit or feature amount extraction unit according to any one of claims 1 to 7”. A recording medium is provided.

[0024]

DESCRIPTION OF THE PREFERRED EMBODIMENTS In the present invention, an attempt was made to optically measure a thin film on a wafer in order to detect a process end point.

Various methods of optically measuring the thickness of a thin film are known, and considerable accuracy is realized even in a method using an interference phenomenon. However, all of them relate to blank film measurement (including multilayer film). The object of the present invention is not only a blank film, but also a substrate (wafer) on which a device pattern (underlying pattern) is formed. Become. In this case, a signal simply predicted from the blank film cannot be obtained.

Therefore, in the present invention, the measurement is performed by using a light source of a multi-wavelength component, irradiating the wafer with the light of the multi-wavelength component, and analyzing the wavelength dependence of the reflected light, that is, the spectral characteristics. Preferably, a white light source is used as the light source of the multi-wavelength component. When a white light source is used, the irradiation may be either white light as it is, or a component obtained by spectrally irradiating the white light over time. Further, the white light source may be a light source that emits light of a plurality of spectra having a relatively narrow half-value width, instead of a light source that normally emits a relatively wide spectrum light. A light source or an ultraviolet light source can be used.

As the irradiation method, a method of irradiating from the polished surface side of the wafer will be described here. However, the irradiating method is not limited to this. It is also possible to adopt a method of performing irradiation from the opposite surface (in this case, there are cases where reflected light is detected and cases where transmitted light is detected).

It is preferable that the spot diameter of the irradiation light be larger than the minimum unit of the pattern. In such a case, the waveform of the spectral characteristic is significantly different from that of the blank film due to a complicated interference effect. Here, the minimum unit of the pattern is, for example, the minimum repetition unit of a pattern having a periodic structure as shown in a one-dimensional direction with respect to the pattern schematically shown in the plan view of FIG.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. FIG. 1 is a schematic view of a CMP polishing apparatus for explaining the present invention. Polishing pad 3 and surface plate 4
The conventional CM shown in FIG. 12 except that a light-transmitting window 5 is provided so as to allow irradiation light to the polished surface of the wafer and reflected signal light to pass through.
Since the operation is the same as that of the P polishing apparatus, the description of the operation of the polishing itself is omitted to avoid redundancy.

The polishing apparatus shown in FIG. 1 polishes the surface to be polished of the wafer 2 by the operation described with reference to the polishing apparatus shown in FIG. In the present invention, the polishing end point measuring device 30 measures the polishing end point during polishing.

The polishing end point measuring device indicated by 30 in FIG. 1 includes a white light source 9, lenses 11 to 13, a beam splitter 10, a light receiving section 6, and a signal processing section 8. Here, as the white light source, a xenon lamp, a halogen lamp, a tungsten lamp, a white LED, or the like can be used. The beam splitter 10 is preferably an optical thin film type of an amplitude division type, and a non-polarization type is preferable in order to reduce an adverse effect on birefringence which a window material generally has on measurement. Further, a computer is preferably used as the signal processing unit 8.

The irradiation light emitted from the white light source is
Through the beam splitter 10 and through the lens 12
Through the light-transmitting window 5 to irradiate the surface of the wafer 2 to be polished. A transparent window material 15 is preferably fitted into the translucent window 5, and polycarbonate, acrylic, or the like is used as a material. The reflected signal light from the wafer 2 passes through the lens 12 again, reflects on the beam splitter 10, passes through the lens 13, and is received by the light receiving unit 6. The light receiving unit 6 sends an optical signal corresponding to the reflected signal light to the signal processing unit 8. The signal processing unit 8 includes a feature amount extraction unit and a logical operation unit.

FIG. 5 shows the configuration of the signal processing unit when a computer is used as the signal processing unit. In FIG. 5, a CPU (central processing unit) 31 is provided inside a computer 30.
The CPU 31 includes an input device 34 including a keyboard and a mouse, a hard disk 36, a memory 37,
The interface boards 33 and 32 are connected. Further, a monitor device is connected to the CPU 31 as necessary.

A CD-ROM drive 35 is connected to the CPU 31. When a CD-ROM 38 in which a signal processing program and its installation program are recorded is inserted into the CD-ROM drive 35, the CPU 31 With this installation program, the signal processing program is expanded and the hard disk 36
In the executable state. When the medium on which the program is recorded is a floppy (registered trademark) disk, a CD-
A floppy disk drive is used instead of the ROM drive 35.

When a computer is used for the signal processing unit,
The feature amount extraction unit performs S2 in FIG. 16 and S32 in FIG.
Also, in S42 of FIG. 18, the CPU 31 corresponds to the “function of extracting a feature amount”, and the logical operation
U31 tunes the membership function (S3), calculates the degree of matching of each feature (S4), calculates the result of each fuzzy rule (S5), and calculates the final result of fuzzy inference (S6). ), Performs defuzzification (S7), and determines whether the value of defuzzification has reached a set value (S8). In FIG. 17, the logical operation unit performs the logical operation based on the logical operation algorithm (S3
3) function of determining whether or not the result of the logical operation satisfies the process end point condition (S34) ".

FIG. 2 shows an example of a waveform of an optical signal. This optical signal is a spectral signal, and the horizontal axis is a spectroscope (not shown).
Channel (117 channels in FIG. 2, wavelength 420 nm)
８００800 nm), and the vertical axis indicates the reflectance. In order to obtain this spectral signal, it is necessary to receive light obtained by splitting the reflected signal light or to use light obtained by splitting white light as irradiation light, but the spectroscope is not shown in FIG. As can be seen from FIG. 1, the platen 4 is rotating, so that the light transmitting window 5 also rotates with respect to the wafer 2 and the irradiation optical axis, and the light transmitting window 5 rotates to the position of the irradiation light. A signal waveform as shown in FIG. 2 is obtained each time it is performed (usually once during rotation of the platen 4). The present invention determines the process end point based on this signal waveform.

As can be seen from FIG. 2, the signal waveform contains many noise elements. Therefore, as preprocessing,
Performs a smoothing process on the signal waveform. FIG. 3 illustrates 16 signal waveforms after performing the smoothing process. In each of the 16 examples, when a wafer having a pattern of a certain type of device is polished, the signal waveforms are continuously obtained each time the platen 4 makes one rotation, and the horizontal axis represents the wavelength and the vertical axis represents the wavelength. Corresponds to reflectivity. A signal waveform acquisition number (hereinafter referred to as a signal number) is displayed at the upper center of the graph of each signal waveform. Therefore, FIG. 3 shows the 32nd (upper left) to 47th (lower right) consecutive signal numbers. In this example, the polishing end point is the time point when the signal number is 45. This 45th signal is the 44th signal before and after it,
As can be seen by comparison with the 46th signal, no clear difference is found between these signals. FIG. 3 does not particularly select a device wafer whose polishing end point is difficult to understand. Generally, the signal change does not show a clear change before and after the polishing end point, and the change is extremely subtle and has ambiguity. We know empirically that it is strong.

As described above, in order to appropriately capture an extremely vague and subtle signal change, the present invention first extracts an appropriate feature from the signal waveform, and finishes polishing based on the change in the feature. Detect points. Further, the polishing end point is detected by a logical operation combining a plurality of these feature amounts.

FIG. 4 shows two signal waveforms for explaining the characteristic amount. These signal waveforms are spectral waveforms, and the lower curve corresponds to signal number 33 in FIG. Here, a maximum value and a minimum value are selected as the feature amounts. On this signal waveform, the maximum value is indicated by ◇, and the minimum value is indicated by +. The maximum value and the minimum value can be calculated (extracted) by performing smoothing differentiation on the signal waveform. The positions (wavelength, reflectance) of the local maximum value and the local minimum value are feature amounts. In this embodiment, the reflectance is used. In extracting these features, the magnitude of the signal waveform is preferably normalized. This standardization is performed independently of changes in the polishing state of the wafer, such as fluctuations in the intensity of the irradiation light source, fluctuations in the transmittance of the optical system including lenses, fluctuations in the light-receiving sensitivity of the light-receiving unit, and fluctuations in the slurry. This is performed to reduce the influence of a fluctuating disturbance component on the signal. As a standardization method, a reference point is designated in a signal waveform, and the size of the signal waveform is corrected so that the size of the reference point is used as a reference value. As the reference point, when the signal waveform is a spectral waveform, the reflectance at a predetermined wavelength in a predetermined spectral range, the maximum value of the reflectance in the predetermined spectral range, or the maximum value in the predetermined spectral range. It is preferable to select one selected from the group of the reflectance, but the present invention is not limited thereto. In the example of FIG. 4, the normalization is performed such that the maximum value of the signal waveform is a predetermined reference value (in this case, 1). Specifically, waveform normalization can be performed by dividing the signal waveform by the largest local maximum value among a plurality of local maximum values. The upper curve in FIG. 4 shows the normalized signal waveform. This normalization of the signal waveform is preferably performed not only when extracting a maximum value or a minimum value as a feature value, but also when extracting any other feature value. It is preferable to perform the processing on the normalized waveform.

Further, when extracting the characteristic amount, it is preferable that after the normalization, the signal waveform is rotationally corrected about the standardized reference point. The reason for doing this is to remove the effects of slurry from the signal waveform. Since the reflected signal light is transmitted through the slurry, the acquired signal waveform contains a component which fluctuates under the influence of scattering or the like due to the slurry. This variation is proportional to the slurry concentration and is wavelength dependent. Since the fluctuation amount is generally larger on the shorter wavelength side, the higher the slurry concentration, the stronger the signal waveform tends to rise to the right. This is shown in FIG. In the state of FIG. 15 (b) and FIG. 15 (c) of this signal waveform,
When gma and SumPB are extracted, their values are different from those in the case of FIG. 15A when there is no slurry, and are dependent on the slurry concentration. In other words, the magnitude of the feature amount is affected by the slurry concentration other than the information inherent in the wafer such as the film thickness, and the measurement accuracy at the process end point is reduced. Therefore, the signal waveform is corrected, and the signal waveform is returned to the state shown in FIG. As a correction method, the signal is rotated around a standardized reference point (point indicated by each X mark in the upper right) of the signal waveform in FIG. 15B or FIG. At this time, the slope is determined by approximating the signal waveform by a linear curve and determining the slope. As a method of correcting the rotation of the signal waveform, there is a method of dividing the signal waveform by the characteristic of the slurry, using the characteristic of the slurry separately measured by a blank mirror or the like as a reference value, in addition to the method by this rotation. Of course, in the case of this second method, normalization is performed after this division.

As other characteristic values, the maximum value or the minimum value in the signal waveform, or the maximum value / minimum value, the maximum value / minimum value, or the pair of adjacent maximum value / minimum value | Maximum value-minimum value |, or an addition value of each | maximum value-minimum value | with respect to a plurality of said maximum value / minimum value pairs, that is, Σ | maximum value-minimum value | The first derivative of the quantity or the second derivative of each feature can be used.

Here, in particular, the feature amount obtained by adding each of the | maximum value−minimum value | to the plurality of the maximum value / minimum value pairs is a difference between the maximum value and the minimum value (Sum OF Pe).
akato Bottom)
Called B. In FIG. 4, SumPB is the sum of elevation differences between peaks and valleys corresponding to adjacent ◇ and + of the normalized waveform, and is ((（1) − (+ 1)) + ((◇ 2) − (+2) +
((◇ 3) − (+ 3)).

Further, the integral value of the signal waveform is simply represented by S
igma. In FIG. 4, Sigma is an area surrounded by the normalized waveform, the wavelength axis, and the vertical axis (reflectance axis). Further, the time derivative of SumPB or Sigma
When the time derivative of “a” is used as a feature value, and this is applied to the case of FIG. 3, the time derivative of SumPB is calculated for the normalized signal (only the original signal is shown in FIG. 3) at each signal number. The slope of SumPB, that is, the difference of SumPB between adjacent signal numbers (for example, between 44 and 45). The time derivative of Sigma is the gradient of Sigma with respect to the normalized signal at each signal number, that is, the difference of Sigma between adjacent signal numbers (for example, between 44 and 45).

By the way, the reflected light from the pattern surface of the semiconductor device wafer reflects the device (
It can be considered as a superposition of light waves from each layer and each part of the (stacked thin film), and the spectral waveform of the reflected signal light resulting from this superposition is due to complicated interference effects (for example, when the film thickness of the top layer is the same). (Even though) it is very different from the blank film. FIG. 13 is an explanatory diagram illustrating the concept of this interference. FIG. 13 shows a cross section of one device wafer. In FIG. 13, 18 is a metal electrode layer, 19 is a dielectric layer,
21 is a lower layer part, 20 is an irradiation light spot, and 10
Reference numerals 0, 200, 300, a, and b denote reflected light waves from each layer and each portion of the (laminated thin film) of the device, respectively. Since the result of the interference of these light waves with each other becomes a reflected signal light, is there.

It is generally not easy to directly calculate the thickness of the object to be measured from the signal waveform obtained from the reflected signal light and determine the polishing state. Further, in addition to the difficulty in analyzing the spectral waveform, there is a problem of disturbance that gives instability to the spectral waveform.

A major first of these disturbances is the slurry. In the case of FIG. 1, the slurry is the slurry adhered to the upper surface of the window plate 15 in the light transmitting window 5. The slurry layer thickness through which the irradiation light and the reflected signal light are transmitted fluctuates irregularly during polishing and the slurry component fluctuates irregularly.
Gives unpredictable noise to the signal waveform.

Secondly, as can be seen from FIG. 1, the rotation of the surface plate 4 causes the light-transmitting window to cross the irradiation light, and every time measurement is performed, the irradiation light spot moves to a position different from the previous irradiation position. Disturbance caused by irradiation and measurement. This disturbance gives unpredictable noise due to inevitable non-uniformity of the remaining film thickness on the wafer and measurement of different pattern types at different positions. [Embodiment 1] As described above, what is handled here is a signal that is difficult to analyze and is affected by disturbance.
For this purpose, the present invention has attempted to extract a plurality of feature amounts capable of capturing a polishing state from a signal waveform and logically operate these feature amounts using fuzzy inference.

In the fuzzy inference of the present embodiment, the following feature values are used. However, other feature value groups can also be used. These feature values are used for experiments or theoretical studies according to the type of wafer. Is selected based on

In this embodiment, SumPB,
Sigma, 1st derivative of SumPB, Sig
ma first derivative, SumPB second derivative,
These are six feature quantities of Sigma's second derivative.

The fuzzy rule is not particularly limited, and is appropriately selected based on experiments or theoretical studies according to the type of wafer and the like. In the present embodiment, the following two rules are used. These rules are linked by "again" in the fuzzy rules. Rule 1: If large, small, small, small, small, negative, and positive, the end points are close. Rule 2: is small, large, large, large, large, or positive,
If it is negative, the end point is far.

Here, each of "large" and "small" in rules 1 and 2 is based on a membership function. The upper figure of FIG. 6 shows SumPB, and the lower figure shows Sig.
FIG. 7 shows a membership function expression of the fuzzy rule.

Here, the membership function is a function indicating the degree of matching (the degree of matching) of the vague words “large” and “small” to the fact that it is large or small, in the fuzzy rule. The fuzzy inference here is based on the Sugano method (Sugeno, M., Industrial app.
lications of fuzzy control, Elsevier Science Pub.C
o., 1985). This membership function is determined in advance for each feature amount based on preliminary experiments, calculation results, and the like.

In the upper part of FIG. 6, the horizontal axis represents the value of SumPB, and the vertical axis represents the degree of matching (degree of matching). Su
When the value of mPB is 1.6 or more, the membership function of “large” is 1 and the membership function of “small” is 0, which means that when the value of SumPB is 1.6 or more, The matching degree between the value of SumPB and “large” is 1, and the matching degree with “small” is 0. When the value of SumPB is 0.8 or less, “small”
Is 1 and the membership function of “large” is 0, which means that the value of SumPB is 0.
When the value is 8 or less, the degree of matching of the value of SumPB with “small” is 1, and the degree of matching with the value of “large” is 0. Further, when the value of SumPB is more than 0.8 and less than 1.6, both the degree of matching with “large” and the degree of matching with “small” take a value of 0 or more and 1 or less. However, it is an area that cannot be said to be “large” or “small”.

In the lower diagram of FIG. 6, Sigma rule 1
The membership functions of “small” and “large” in Rule 2 are shown, and their meanings may be interpreted in the same manner as in SumPB.

Next, in FIG. 7, (1), (2), (3),
(4), (5) and (6) are SumPB and Sig described above, respectively.
ma, the first derivative of SumPB (SumPB-Dif
f), Sigma's first derivative (Sigma-Dif)
f), the second derivative of SumPB (SumPB-Dif)
f2) and Sigma's second derivative (Sigma-
Diff2) is a membership function. Of the upper and lower two columns, the upper column is for rule 1 and the lower column is for rule 2. Where (1) and (2)
Shows the membership function shown in FIG. 6 divided into rule 1 and rule 2 and reduced and displayed.
In the membership functions (3), (4), (5), and (6),
The horizontal axis represents the value of each feature value, and the vertical axis represents the degree of matching (0 to 1) for each of Rule 1 and Rule 2. Also, each straight line parallel to each vertical axis of each membership function is an input value of each feature amount of,,, and for a certain signal number, and is 2.50, 75, 0.12,. 5
0, -1.00, and 1.00. Each intersection between each of these straight lines and each of the membership functions is the degree of matching of each feature.

For rule 1, (1), (2), (3), (4),
The matching degrees of (5) and (6) are 1, 1, 0.60, 0.
75, 1, and 1, and Rule 1 takes the logical product of these. In this embodiment, an algebraic product is taken as a logical product.
The result of rule 1 is 1 × 1 × 0.60 × 0.75 × 1 ×
1 = 0.45. The result of this rule 1 is shown in FIG.
(A) shows the degree of matching with respect to 1 when the polishing end point is set to 1, which is 0.45.
Is the membership function of the result.

Further, (1), (2), (3),
(4), (5), and (6) are 0, 0, 0.4,
0.25, 0, and 0, and Rule 2 takes the logical sum of these. Algebraic sum is used as the logical sum, and the result of rule 2 is 0 + 0 + 0.4 + 0.25 + 0 + 0− (0.4
× 0.25) = 0.55. The result of this rule 2 is shown in FIG. 7 (b), where the degree of coincidence with respect to 0 when the completely unfinished point of polishing is 0 is 0.55, which is a member of the result of rule 2. It is a ship function.

Next, if the result of rule 1 and the result of rule 2 are expressed together in FIG. 7 (c), the result of rule 1 and the result of rule 2 are linked by "again". This is the end result of the fuzzy inference, again a membership function.

Next, it is preferable to perform defuzzification to extract the essence from the membership function shown in FIG. 7C. This defuzzification method preferably obtains the center of gravity of the membership function of the final result, but is not limited to this method. When obtaining from the center of gravity, the following equation is used to calculate the center of gravity = 1 × 0.45 + 0 × 0.55 / 0.45 + 0.5
5 = 0.45 0.45, which is the value at the time of polishing (signal number)
Is used as the end point evaluation value in

In this fuzzy inference, the closeness of the polishing end point,
That is, the end point evaluation value is indicated by a value in the range from 0 to 1,
It is known in advance that the time when the value reaches 0.9 or more is the polishing end point. FIG. 8 shows how the end point evaluation value changes.
In FIG. 8, the horizontal axis represents the signal number, and the vertical axis represents the end point evaluation value (0
Since the end point evaluation value is 0.9 or more when the signal number is 33, this can be determined as the polishing end point, and the polishing end point signal can be output.

Next, a detailed description will be given of a change in a feature value extracted from a signal waveform, which is a basis for performing the fuzzy inference. FIG. 9 shows an example of the change of SumPB. The horizontal axis indicates the signal number, which corresponds to the number of times of polishing. The dotted line indicates the value of SumPB, and the solid line indicates the value of the moving average of SumPB. FIG. 10 shows an example of the change in Sigma. The meanings of the dotted line, solid line, and horizontal axis are the same as in FIG. Su
For both mPB and Sigma, the input to the fuzzy rule uses the value of the moving average.

Rules 1 and 2 for fuzzy inference described above
Among them, the rule of “large” and “small” in SumPB is a rule for determining the magnitude of the change in the signal waveform, and the rule of “large” and “small” in Sigma is the rule of the entire signal waveform. Since this is a rule for determining the degree of the size, this part can be said to be a quantitative rule for feature amount determination. In addition, 1 of SumPB
Derivative, Sigma's first derivative, SumP
The second derivative of B and the second derivative of Sigma are rules for finding local maxima and minima in the curves of SumPB and Sigma in FIGS. 9 and 10, and a qualitative rule for grasping the shape of the curves. I can say.

Regarding the quantitative part and the value, the value greatly varies depending on the type of wafer and the type and state of slurry, and therefore the value of SumPB and Sigma with the progress of polishing are large.
It is preferable to perform tuning to shift the membership function sideways during measurement according to the change in the value of. As a reference value for tuning, it is preferable to use an average value of the feature amounts from the start of polishing to the time of measurement. Each solid line parallel to the horizontal axis in the graphs of FIGS. 9 and 10 indicates the average value. Thus, for example, at each measurement stage during polishing, Sum
After calculating the average value of PB, Sum
Tuning is performed, for example, by shifting the membership function shown in the upper part of FIG. 6 horizontally so that both the degree of matching with the PB membership function and the degree of matching with the “large” and “small” become 0.5. Tuning of the Sigma membership function is also performed by shifting the membership function shown in the lower diagram of FIG. 6 horizontally, as in the case of SumPB.

By this tuning, SumPB and S
Even when the value of igma changes due to fluctuation of the slurry, etc., the membership function can be appropriately selected.
As described above, in the present invention, two or more features are extracted from a signal waveform, and a logic operation is performed on these features using fuzzy inference, so that the measurement target is a substrate having a device pattern. Even if there is disturbance due to fluctuations in the slurry or the measurement position, the polishing end point can be detected with high accuracy and simultaneously with polishing.

When the signal processing operation of the signal processing device in the above description is performed by a computer, the operation is shown in FIG. Hereinafter, the operation of the signal processing device will be described with reference to step numbers.

First, when the signal processing device is turned on, the CPU 31 in FIG. 5 acquires an optical signal (S1). This optical signal is obtained at each sampling period interval. Next, C
The PU 31 extracts a feature amount from the optical signal (S2). Prior to the extraction of the feature, the feature is selected (S10). This selection may be made manually or automatically selected in accordance with the type of wafer or the like.

Next, the CPU 31 tunes the membership function (S3). Prior to this S3,
A membership function has been determined (S11). This determination may be made manually or automatically selected in accordance with the type of wafer or the like.

Next, the CPU 31 calculates each degree of coincidence with respect to the input value of each feature amount (S4). Next, CPU3
1 calculates the result of the fuzzy rule (S5).

Prior to the step of S5, a fuzzy rule has been determined (S12). This decision method is
Depending on the wafer type or the like, it may be performed manually or determined automatically.

Next, the CPU 31 calculates the final result of the fuzzy inference by combining the results of the respective fuzzy rules (S6). Next, the CPU 31 defuzzifies the final result of the fuzzy inference (S7).

Next, the CPU 31 determines whether or not the value of the defuzzification has reached a value preset as a process end point (S8). Prior to S8, the value of the process end point is set (S13). This setting may be performed manually or determined automatically according to the type of wafer.

If NO in S8, the CPU 31 performs a process on an optical signal sampled and acquired next. If YES in S8, CPU 31 outputs a process end point signal (S9).

In this embodiment, the extracted feature amount is used.
The end point of polishing is detected by fuzzy inference.
Even if there is a disturbance in the signal,
Even if there is a high accuracy and stable detection of the process end point,
Can perform one or both of simultaneous detection. [Embodiment 2]
In the measurement of the first embodiment, a logical operation of two or more feature amounts is performed.
Detection of polishing end point using fuzzy inference
However, when the disturbance of the signal waveform is small, fuzzy
If the required accuracy can be obtained without using inference,
Is because using fuzzy inference complicates logical operations
The cost may increase. In these cases,
Do not use fuzzy inference. For example, the fuzzy
SumPB is a threshold instead of fuzzy rule 1
S₁Larger and Sigma is threshold S_TwoSmaller than
And the first derivative of SumPB has a threshold S_ThreeLess than
And Sigma's first derivative is threshold S_Fourthan
Small and the second derivative of SumPB is a negative value
Polishing if the second derivative of Sigma is positive
Each of these features is defined as an end point.
The logical expression that the polishing end point is reached when all the conditions are satisfied
Algorithm. Where S ₁, S
_Two, S_Three, S_FourIs a constant value determined for each wafer.

When the signal processing operation of the signal processing apparatus in the above description is performed by a computer, the operation is shown in FIG. Hereinafter, the operation of the signal processing device will be described with reference to step numbers.

First, when the signal processing device is turned on, the CPU
31 acquires an optical signal (S31). This optical signal is
It is acquired at each sampling cycle interval. Next, CPU
31 extracts a feature amount from the optical signal (S32). Prior to the extraction of the feature, the feature is selected (S
36). This selection may be made manually or automatically selected in accordance with the type of wafer or the like.

Next, the CPU 31 performs a logical operation (S
33). Prior to S33, a logical operation algorithm is determined (S37). This determination may be made manually or automatically selected in accordance with the type of wafer or the like.

Next, the CPU 31 determines whether or not the result of the logical operation satisfies the process end point condition (S34).
If NO in S34, the CPU 31 performs a process on the optical signal sampled and acquired next.

If YES in S34, CPU 31 outputs a process end point signal (S35). In this embodiment,
Since the polishing end point is detected by a logical operation using the extracted feature quantity, even if there is a disturbance in the signal or the wafer on which the pattern is formed, it does not reach the case of the first embodiment, One or both of high-precision and stable detection of the process end point, or simultaneous detection can be performed. [Embodiment 3] In the measurements of Embodiments 1 and 2 described above, two or more feature amounts are selected from the signal waveform, and the polishing end point is detected based on these logical operations. Depending on the type), there is a case where a logical operation is rather unfavorable, or a case where performing a logical operation becomes a problem in terms of cost. In this case, only one feature value is selected, and the change is detected. The selected feature amount is | maximum value−minimum value | for a pair of adjacent maximum value / minimum value in a signal waveform (in this case, a spectral waveform), or each | maximum value of a plurality of the maximum value / minimum value pairs. Either the sum of the minimum values | or the integral value of the signal waveform is preferable. In this case, the measurement is simplified in polishing the wafer having the pattern.

When the signal processing operation of the signal processing device in the above description is performed by a computer, the operation is shown in FIG. Hereinafter, the operation of the signal processing device will be described with reference to step numbers.

First, when the signal processing device is turned on, the CPU
31 acquires an optical signal (S41). This optical signal is
It is acquired at each sampling cycle interval. Next, CPU
31 extracts a feature amount from the optical signal (S42). Prior to the extraction of the feature, the feature is selected (S
45). This selection may be made manually or automatically selected in accordance with the type of wafer or the like.

Next, the CPU 31 determines whether or not the feature amount has reached a set value (S43). Prior to S43, the value of the process end point is set (S46). This setting may be performed manually or automatically according to the type of the wafer.

If NO in S43, the CPU 31 performs a process on an optical signal sampled and acquired next. If YES in S43, CPU 31 outputs a process end point signal (S44).

FIG. 19 shows an example measured by the present measuring method. FIG. 19 shows a TEG (TestElement Gr).
oove) Sigma for the signal number when Sigma and SumPB are selected as the feature amounts for the pattern
And changes in SumPB. In the case of the present embodiment, the point of time of signal number 50 corresponds to the polishing end point, and at this point, Si
Since both gma and SumPB change rapidly, the polishing end point can be detected by capturing this timing. In this case, the feature amount (in this case, S
igma and SumPB) by performing the differentiation once or twice, so that the detection of the polishing end point is further facilitated.
It is preferable to use one time and two times differentiation of a signal together.

In this embodiment, since the polishing end point is detected based on the change in the extracted feature value, it is not necessary to use a logical operation algorithm or fuzzy inference. Even in this case, one or both of the end points of the process can be simply and accurately detected, or the simultaneous detection can be performed. Further, depending on the type of device pattern, detection can be performed with higher accuracy than in the case of performing a logical operation.

The measuring apparatus using the measuring method described in the first, second, and third embodiments is provided in a polishing apparatus or the like, and is used for measuring a process state. [Embodiment 4] This embodiment relates to a method of manufacturing a semiconductor device using the polishing apparatus of the present invention.

FIG. 11 is a flowchart showing a semiconductor device manufacturing process. After the semiconductor device manufacturing process is started, first, in step S200, an appropriate processing step is selected from the following steps S201 to S204. Steps S201 to S20 according to the selection
Go to any of 4

Step S201 is an oxidation step for oxidizing the surface of the silicon wafer. Step S202 is CV
D to form an insulating film on the silicon wafer surface
This is a VD process. Step S203 is an electrode film forming step of forming an electrode film on a silicon wafer by a process such as vapor deposition. Step S204 is an ion implantation step of implanting ions into the silicon wafer.

After the CVD step or the electrode film forming step, the flow advances to step S209 to determine whether or not to perform the CMP step. If not, the process proceeds to step S206; otherwise, the process proceeds to step S205. Step S2
Reference numeral 05 denotes a CMP step, in which a polishing apparatus of the present invention is used to planarize an interlayer insulating film and to form a damascene by polishing a metal film on the surface of a semiconductor device.

After the CMP step or the oxidation step, the process proceeds to step S206. Step S206 is a photolithography step. In the photolithography process, a resist is applied to a silicon wafer, a circuit pattern is printed on the silicon wafer by exposure using an exposure device, and the exposed silicon wafer is developed. Further, in the next step S207, portions other than the developed resist image are removed by etching.
Thereafter, the resist is stripped, and the etching step is performed to remove the unnecessary resist after the etching.

Next, it is determined in step S208 whether all necessary processes have been completed, and if not, step S20
Returning to 0, the above steps are repeated to form a circuit pattern on the silicon wafer. If it is determined in step S208 that all steps have been completed, the process ends.

In the semiconductor device manufacturing method according to the present invention, since the polishing apparatus according to the present invention is used in the CMP step, the accuracy of detecting the polishing end point in the CMP step is improved, and the yield in the CMP step is reduced. improves.
As a result, there is an effect that a semiconductor device can be manufactured at a lower cost than a conventional semiconductor device manufacturing method.

The present invention can be used in a CMP step of a semiconductor device manufacturing process other than the semiconductor device manufacturing process shown in FIG. The semiconductor device according to the present invention is manufactured by the semiconductor device manufacturing method according to the present invention. As a result, a semiconductor device can be manufactured with higher quality and at lower cost than the conventional semiconductor device manufacturing method, and the manufacturing cost of the semiconductor device can be reduced.

The invention of the first, second, third, and fourth embodiments has been described above. However, a function that enables any two or more measurement methods selected from the signal processing methods of the first, second, and third embodiments. May be incorporated in one measuring device, and any one of the functions may be selected and used for the measurement. This makes it possible to select an optimal measurement method for the type of wafer and polishing conditions.

The present invention is not limited to the case where the measurement is performed through the light-transmitting window as shown in FIG. 1, but also enables the polishing head to swing in addition to the rotation so that the wafer protrudes from the polishing pad, and the protruding portion is exposed to light. Irradiation and measurement. In this case, no translucent window is required. Further, in a polishing apparatus in which the polishing pad is smaller than the wafer, the measurement can also be performed on a portion where the wafer protrudes from the polishing pad and is exposed.

The present invention is not limited to the polishing end point,
It can also be used for detecting the end point of other removal processes such as ion etching and the like, and the film formation process such as CVD and sputtering. Further, the terminating point of the process referred to here includes not only the terminating point of the step in a general thin film removing step but also the terminating point of an intermediate step such as the timing at which the removing step proceeds to a different material layer.

Although the measuring apparatus shown in FIG. 1 emits light from the pattern surface side of the semiconductor device, the light can be emitted from the back side of the wafer. In this case, the light source requires a multi-wavelength component light source in the infrared region.

Although the present invention has been described with reference to the drawings, the scope of the present invention is not limited to the ranges shown in these drawings, and the present invention is limited to the above description. Not even a thing.

[0098]

According to the present invention, since the feature quantity extracted from the signal waveform is logically operated, the polishing layer can be formed even if there is a pattern on the substrate or there is some disturbance in the signal waveform. Even when there is no clear change, one or both of the high-precision detection of the process end point and the simultaneous detection can be performed. [Claim 2] In the present invention, since a characteristic amount that appropriately changes in accordance with a change in the thickness of a thin film is selected, one or both of high-precision detection of the process end point and / or simultaneous detection of the process end point can be performed with higher accuracy. it can. [Claim 3] In the present invention, since fuzzy inference is used for a logical operation, one or both of high-precision detection and / or simultaneous detection of a process end point can be performed even if there is a large disturbance in a signal waveform. [Claim 4] In the present invention, the membership function is tuned during the measurement, so that a membership function suitable for the magnitude of the extracted feature can be set at any time. One or both of high-precision detection of points and / or simultaneous detection can be performed. [Claim 5] In the present invention, a process end point is measured by monitoring a change in a feature amount that reflects a process state extremely well, and a logical operation is not performed, so that the process end point is detected simply and with high accuracy. One or both of the simultaneous detections can be performed. Further, it is most suitable depending on the type of the device pattern. [Claim 6] According to the present invention, since the characteristic amount is extracted from the normalized waveform, even if there is a disturbance, one or both of the end points of the process can be detected simply and with high accuracy, or the simultaneous detection can be performed. [Claim 7] In the present invention, even if the signal waveform is tilted due to an abrasive or the like, the feature amount is extracted from the waveform obtained by rotation-correcting the signal waveform. You can do one or both. [Claim 8] In the present invention, since the feature quantity extracted from the signal waveform is logically operated, the polishing layer does not change clearly even if there is a pattern on the substrate or the signal waveform has some disturbance. However, one or both of high-precision detection and simultaneous detection of the process end point can be performed. [Claim 9] In the present invention, a process end point is measured by monitoring a change in a feature amount that reflects a process state extremely well, and a logical operation is not performed. Therefore, the process end point is detected simply and with high accuracy. One or both of the simultaneous detections can be performed. Further, it is most suitable depending on the type of the device pattern. [Claim 10] In the present invention, since the measuring apparatus according to any one of claims 1 to 7 is provided, a wafer having a device pattern can be polished with high precision or with good yield. [Claim 11] In the present invention, a semiconductor device is polished by the polishing apparatus according to claim 10, so that a semiconductor device can be manufactured with high quality or at low cost. [Claim 12] According to the present invention, it is possible to realize the "feature amount extraction unit and logical operation unit according to any one of claims 1 to 7" on a computer.

[Brief description of the drawings]

FIG. 1 is a schematic diagram of a CMP polishing apparatus of the present invention.

FIG. 2 shows an example of a signal waveform (spectral waveform).

FIG. 3 is an example of continuous display of a signal waveform (spectral waveform).

FIG. 4 is a diagram showing a relationship between a signal waveform, a normalized signal waveform, and a feature amount;

FIG. 5 is a diagram of a signal processing unit using a computer.

FIG. 6 shows examples of membership functions of SumPB and Sigma.

FIG. 7 is an example of a membership function expression of a fuzzy rule.

FIG. 8 is a graph of end point evaluation values with respect to signal numbers.

FIG. 9 is a graph of the signal number of SumPB.

FIG. 10 is a graph of Sigma signal numbers.

FIG. 11 is a flowchart illustrating an example of a semiconductor device manufacturing process.

FIG. 12 is a schematic view of a conventional CMP polishing apparatus.

FIG. 13 shows the relationship between the reflected light wave from each layer and each part of the device and the irradiation light spot.

FIG. 14 is an explanatory diagram of a minimum unit of a pattern.

FIG. 15 is a diagram showing an outline of rotation of a signal waveform.

FIG. 16 is a flowchart of an embodiment of signal processing based on fuzzy inference.

FIG. 17 is a flowchart of an embodiment of signal processing by a logical operation.

FIG. 18 is a flowchart of an embodiment of signal processing based on a change in a feature value.

FIG. 19: Changes in Sigma and SumPB

[Explanation of symbols]

1 Polishing head 2 Substrate (wafer) 3 Polishing pad 4 Surface plate 5 Light transmitting window 6 Light receiving unit 7 Irradiation light and reflected light 8 Signal processing unit (personal computer) 9 White light source 10 Beam splitter 11-13 Lens 15 Transparent window material 16 Abrasive (slurry) supply mechanism 17 Abrasive (slurry) 18 Metal electrode layer 19 Dielectric layer 20 Irradiation light spot 21 Lower layer portion 30 Polishing end point measuring device a, b, 100, 200, 300 Reflected light wave ω _H Polishing head show the rotation of the ω _T plate indicating the rotation

────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] September 12, 2001 (2001.9.1)
2)

[Procedure amendment 1]

[Document name to be amended] Statement

[Correction target item name] Full text

[Correction method] Change

[Correction contents]

[Document Name] Statement

Patent application title: Process end point measuring apparatus, measuring method, polishing apparatus, semiconductor device manufacturing method, and recording medium recording signal processing program

[Claims]

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a process for manufacturing a semiconductor device such as an LSI, for example, and detects an end point of the process in a process of forming a film on a semiconductor wafer or a process of polishing a thin film on the wafer. The present invention relates to a measuring apparatus, a measuring method, a polishing apparatus, a semiconductor device manufacturing method, and a recording medium on which a measuring method program is recorded.

[0002]

2. Description of the Related Art High density semiconductor devices have continued to evolve without showing any limitations.
Methods are being developed. One of them is a multilayer wiring, and the technical problems associated with this include flattening a global (relatively large area) device surface on a semiconductor wafer and wiring between upper and lower layers.

[0003] Shortening the exposure wavelength in the lithography process,
Furthermore, high NA (numerical aperture)
Considering the reduction in the depth of focus during exposure accompanying the above, there is a great demand for the accuracy of flattening the interlayer layer at least in the range of the exposure area. There is also a great demand for so-called inlays (plugs, damascenes), which are burying of metal electrode layers in order to realize multilayer wiring, and in this case, it is necessary to remove and flatten an extra metal layer after lamination of the metal layers. . A polishing process called CMP is drawing attention as an efficient planarization technique for these large areas. CMP (Ch
emical Mechanical Polishing or Planarization)
Is a process that removes the surface layer of a wafer by using both physical polishing and chemical polishing (dissolution with a polishing agent solution), and is the leading global flattening and electrode forming technology. Has become a candidate. Specifically, using an abrasive called a slurry in which abrasive particles (typically silica, alumina, cerium oxide, etc.) are dispersed in a soluble solvent of the object to be polished, such as an acid or an alkali, and using an appropriate polishing pad Then, the polishing is advanced by pressing the wafer surface and rubbing by relative motion. By making the pressure and the relative movement speed uniform over the entire surface of the wafer, uniform polishing within the surface becomes possible.

FIG. 12 is a schematic view of a conventional CMP polishing apparatus. The wafer 2 mounted on the polishing head 1 is pressed against the polishing pad 3 while rotating at an angular velocity ωH. The platen 4 to which the polishing pad is fixed rotates at an angular velocity ωT.
An abrasive supply mechanism 16 is provided between the wafer 2 and the polishing pad 3.
(Slurry) 17 is supplied from the
The polished surface of the wafer 2 is polished by the chemical action and the physical action of the polishing pad 3. The polishing rate v at an arbitrary point in the surface of the wafer 2 is given by: v = rC where rC is the distance from the center of the platen 4 to the center of the polishing head 1 and rH is the distance from the center of the polishing head 1 to the polishing point. ωT −rH ・ (ωH
−ωT), the wafer 2 when ωH = ωT
The polishing rate becomes constant irrespective of the position inside.

[0005] One of the major requirements of this process is as follows.
There is detection of the end point of the polishing process. In particular, the detection of the polishing end point (in-situ) during the polishing process is greatly demanded in order to improve the process efficiency.

As this detection method, a general film thickness measuring device is often used for detecting the end point of the polishing process. A minute blank portion (where no device pattern is present) of the wafer cleaned after the process is selected as a measurement location, and detection and measurement are performed by various methods.

In the polishing and flattening step, as a faster monitoring method, a method of detecting a fluctuation in friction when polishing proceeds to a layer different from the target polishing layer by a change in motor torque of wafer rotation or pad rotation. is there.

Another method is to irradiate a laser beam to a wafer surface, track the time variation of reflected light intensity using optical interference, and measure the film thickness. In many cases, the end point is determined when the intensity changes over time and reaches a predetermined value.However, due to the uncertainty depending on the device pattern of the wafer, errors due to the measurement position, and the effects of signal noise, the process ends. It is pointed out that it is difficult to judge points clearly.

[0009]

There are various methods for detecting the end point in the above-described CMP process, but no definitive method has yet been found.

For example, measurement with a film thickness measuring instrument can obtain sufficient accuracy and reliable data, but the apparatus itself becomes large-scale, measurement takes time, and feedback to the process is required. Become slow.

Although the method of detecting the process end point by the motor torque is simple and high-speed, it is effective only when detecting that the layer has clearly changed to a different type as the process end point, and is more accurate. Is not enough.

Furthermore, the method of irradiating a laser beam onto a wafer surface combines the uncertainty and error of the measurement position depending on the device pattern type of the wafer, and the influence of signal noise caused by slurry and the like. It has been pointed out that it is difficult to clearly determine the process end point due to signal disturbance.

[0013] The present invention solves the above-mentioned problems, and enables the in-situ measurement of the polishing end point even if the signal is disturbed and the polishing layer does not change to a clearly different type. Provided are a measuring device, a measuring method, a polishing device, and a recording medium on which a semiconductor device manufacturing method and a measuring method are recorded.

[0014]

Means for Solving the Problems In order to solve the above problems, the present invention firstly sets a step of forming an insulating film or a metal electrode film on a substrate, or a step end point in a step of removing the film. The substrate surface is irradiated with light, and the reflected signal light or
Is a measuring device for measuring from a signal waveform obtained by detecting transmitted signal light , wherein the signal waveform is a spectral waveform, and a feature value extracting unit for extracting two or more feature values from the signal waveform; One or more performs logical operation using the feature quantity comprises a logical operation unit determines process end point, and each feature quantity
Correspond to adjacent maximum / minimum value pairs in the signal waveform.
| Maximum value−minimum value | and a plurality of the maximum value / minimum value pairs
Sum of each | maximum value−minimum value |
Integral of the shape and one and two time derivatives of each of the values
And a group of positive and negative signs of the time derivative,
A measuring device characterized by being one selected from the group consisting of:

Second, an insulating film or a metal electrode on the substrate
End of process in film formation process or film removal process
At the end point, the substrate surface is irradiated with light, and the reflected signal light or
Is measured from the signal waveform obtained by detecting the transmitted signal light
A measurement device, wherein the signal waveform is a spectral waveform,
A feature extraction unit that extracts two or more features from a signal waveform
And fuzzy inference using the two or more features
Logical operation unit that performs the specified logical operation and determines the process end point
And a measuring device characterized by comprising:

Third, the fuzzy inference used
To the value calculated from the feature quantity
Tuned during the measurement.
2. A measuring device according to item 2.

Fourthly, the step of forming an insulating film or a metal electrode film on the substrate or the end point of the step of removing the film is performed by irradiating the substrate surface with light and reflecting the reflected signal light or the reflected signal light.
Is a measuring device that measures from a change in a feature value extracted from a signal waveform obtained by detecting a transmitted signal light , and includes a feature value extraction unit that extracts a feature value from the signal waveform, and the signal waveform is A spectral waveform, and the feature amount is
| For the adjacent maximum / minimum value pair in the signal waveform
A measuring device is provided which is a maximum value-minimum value |, or an addition value of each | maximum value-minimum value | for a plurality of the maximum value / minimum value pairs, or an integrated value of the signal waveform.

[0018] Fifth, according to claim 1-4, characterized by extracting the feature amount of the signal waveform from the normalized waveform
A measurement device according to any one of the preceding claims. Sixth, the measuring device according to any one of claims 1 to 5 , wherein the characteristic amount is extracted from a waveform obtained by rotating and correcting the signal waveform.

Seventh, the step of forming an insulating film or a metal electrode film on the substrate or the end point of the step of removing the film is performed by irradiating the substrate surface with light and reflecting the reflected signal light or the reflected signal light.
Is a measuring method for measuring from a signal waveform obtained by detecting a transmitted signal light, and extracting two or more characteristic amounts from the signal waveform, and performing a logical operation using the two or more characteristic amounts.
E ingredients and step of determining process end point, was subjected to, each JP
The characteristic amount is determined by the pair of the adjacent maximum value and minimum value in the signal waveform.
| Maximum value-minimum value | and a plurality of the maximum value / minimum value
Value of each | maximum value−minimum value |
Signal waveform integral value and one time and two time derivative of each value
A group of coefficients and a group of positive and negative signs of the time derivative
Provided is a measurement method characterized by being one selected from a group .

Eighth, the step of forming an insulating film or a metal electrode film on the substrate, or the end point of the step of removing the film is performed by irradiating the surface of the substrate with light and reflecting the reflected signal light or
Is a measurement method for measuring by a change in a characteristic amount extracted from a signal waveform obtained by detecting a transmitted signal light , wherein the signal waveform is a spectral waveform, and the characteristic amount is | Maximum value-minimum value | for an adjacent local maximum value / minimum value pair, or an added value of each | maximum value-minimum value | And a measuring method characterized by the following.

Here, for the adjacent maximum value / minimum value pair,
| Maximum value-minimum value |
Indicates the absolute value of the value obtained by subtracting the minimum value from the maximum value.

Ninth, a holding unit for holding the substrate, a polishing body, and the measuring device according to any one of claims 1 to 6 , comprising an abrasive between the substrate and the polishing body. When a load is applied between the substrate and the polishing body in a state where the substrate is interposed and a substrate is polished by applying a relative motion between the two, the end point of the process can be measured. The present invention provides a polishing apparatus characterized by the following.

[0023] Tenth, to provide a semiconductor device manufacturing method characterized by comprising the step of polishing a surface of a semiconductor wafer using a polishing apparatus according to claim 9, wherein. Eleventh,
A machine-readable recording medium storing a signal processing program for causing a computer to function as a “feature amount extraction unit and a logical operation unit or a feature amount extraction unit according to any one of claims 1 to 6 ”. .

[0024]

DESCRIPTION OF THE PREFERRED EMBODIMENTS In the present invention, an attempt was made to optically measure a thin film on a wafer in order to detect a process end point.

Various methods of optically measuring the thickness of a thin film are known, and considerable accuracy is realized even in a method using an interference phenomenon. However, all of them relate to blank film measurement (including multilayer film). The object of the present invention is not only a blank film, but also a substrate (wafer) on which a device pattern (underlying pattern) is formed. Become. In this case, a signal simply predicted from the blank film cannot be obtained.

Therefore, in the present invention, the measurement is performed by using a light source of a multi-wavelength component, irradiating the wafer with the light of the multi-wavelength component, and analyzing the wavelength dependence of the reflected light, that is, the spectral characteristics. Preferably, a white light source is used as the light source of the multi-wavelength component. When a white light source is used, the irradiation may be either white light as it is, or a component obtained by spectrally irradiating the white light over time. Further, the white light source may be a light source that emits light of a plurality of spectra having a relatively narrow half-value width, instead of a light source that normally emits a relatively wide spectrum light. A light source or an ultraviolet light source can be used.

As the irradiation method, a method of irradiating from the polished surface side of the wafer will be described here. However, the irradiating method is not limited to this. It is also possible to adopt a method of performing irradiation from the opposite surface (in this case, there are cases where reflected light is detected and cases where transmitted light is detected).

It is preferable that the spot diameter of the irradiation light be larger than the minimum unit of the pattern. In such a case, the waveform of the spectral characteristic is significantly different from that of the blank film due to a complicated interference effect. Here, the minimum unit of the pattern is, for example, the minimum repetition unit of a pattern having a periodic structure as shown in a one-dimensional direction with respect to the pattern schematically shown in the plan view of FIG.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. FIG. 1 is a schematic view of a CMP polishing apparatus for explaining the present invention. Polishing pad 3 and surface plate 4
The conventional CM shown in FIG. 12 except that a light-transmitting window 5 is provided so as to allow irradiation light to the polished surface of the wafer and reflected signal light to pass through.
Since the operation is the same as that of the P polishing apparatus, the description of the operation of the polishing itself is omitted to avoid redundancy.

The polishing apparatus shown in FIG. 1 polishes the surface to be polished of the wafer 2 by the operation described with reference to the polishing apparatus shown in FIG. In the present invention, the polishing end point measuring device 30 measures the polishing end point during polishing.

The polishing end point measuring device indicated by 30 in FIG. 1 includes a white light source 9, lenses 11 to 13, a beam splitter 10, a light receiving section 6, and a signal processing section 8. Here, as the white light source, a xenon lamp, a halogen lamp, a tungsten lamp, a white LED, or the like can be used. The beam splitter 10 is preferably an optical thin film type of an amplitude division type, and a non-polarization type is preferable in order to reduce an adverse effect on birefringence which a window material generally has on measurement. Further, a computer is preferably used as the signal processing unit 8.

The irradiation light emitted from the white light source is
Through the beam splitter 10 and through the lens 12
Through the light-transmitting window 5 to irradiate the surface of the wafer 2 to be polished. A transparent window material 15 is preferably fitted into the translucent window 5, and polycarbonate, acrylic, or the like is used as a material. The reflected signal light from the wafer 2 passes through the lens 12 again, reflects on the beam splitter 10, passes through the lens 13, and is received by the light receiving unit 6. The light receiving unit 6 sends an optical signal corresponding to the reflected signal light to the signal processing unit 8. The signal processing unit 8 includes a feature amount extraction unit and a logical operation unit.

FIG. 5 shows the configuration of the signal processing unit when a computer is used as the signal processing unit. In FIG. 5, a CPU (central processing unit) 31 is provided inside a computer 30.
The CPU 31 includes an input device 34 including a keyboard and a mouse, a hard disk 36, a memory 37,
The interface boards 33 and 32 are connected. Further, a monitor device is connected to the CPU 31 as necessary.

A CD-ROM drive 35 is connected to the CPU 31. When a CD-ROM 38 in which a signal processing program and its installation program are recorded is inserted into the CD-ROM drive 35, the CPU 31 With this installation program, the signal processing program is expanded and the hard disk 36
In the executable state. When the medium on which the program is recorded is a floppy disk, a floppy disk drive is used instead of the CD-ROM drive 35.

When a computer is used for the signal processing unit,
The feature amount extraction unit performs S2 in FIG. 16 and S32 in FIG.
Also, in S42 of FIG. 18, the CPU 31 corresponds to the “function of extracting a feature amount”, and the logical operation
U31 tunes the membership function (S3), calculates the degree of matching of each feature (S4), calculates the result of each fuzzy rule (S5), and calculates the final result of fuzzy inference (S6). ), Performs defuzzification (S7), and determines whether the value of defuzzification has reached a set value (S8). In FIG. 17, the logical operation unit performs the logical operation based on the logical operation algorithm (S3
3) function of determining whether or not the result of the logical operation satisfies the process end point condition (S34) ".

FIG. 2 shows an example of a waveform of an optical signal. This optical signal is a spectral signal, and the horizontal axis is a spectroscope (not shown).
Channel (117 channels in FIG. 2, wavelength 420 nm)
８００800 nm), and the vertical axis indicates the reflectance. In order to obtain this spectral signal, it is necessary to receive light obtained by splitting the reflected signal light or to use light obtained by splitting white light as irradiation light, but the spectroscope is not shown in FIG. As can be seen from FIG. 1, the platen 4 is rotating, so that the light transmitting window 5 also rotates with respect to the wafer 2 and the irradiation optical axis, and the light transmitting window 5 rotates to the position of the irradiation light. A signal waveform as shown in FIG. 2 is obtained each time it is performed (usually once during rotation of the platen 4). The present invention determines the process end point based on this signal waveform.

As can be seen from FIG. 2, the signal waveform contains many noise elements. Therefore, as preprocessing,
Performs a smoothing process on the signal waveform. FIG. 3 illustrates 16 signal waveforms after performing the smoothing process. In each of the 16 examples, when a wafer having a pattern of a certain type of device is polished, the signal waveforms are continuously obtained each time the platen 4 makes one rotation, and the horizontal axis represents the wavelength and the vertical axis represents the wavelength. Corresponds to reflectivity. A signal waveform acquisition number (hereinafter referred to as a signal number) is displayed at the upper center of the graph of each signal waveform. Therefore, FIG. 3 shows the 32nd (upper left) to 47th (lower right) consecutive signal numbers. In this example, the polishing end point is the time point when the signal number is 45. This 45th signal is the 44th signal before and after it,
As can be seen by comparison with the 46th signal, no clear difference is found between these signals. FIG. 3 does not particularly select a device wafer whose polishing end point is difficult to understand. Generally, the signal change does not show a clear change before and after the polishing end point, and the change is extremely subtle and has ambiguity. We know empirically that it is strong.

As described above, in order to appropriately capture an extremely vague and subtle signal change, the present invention first extracts an appropriate feature from the signal waveform, and finishes polishing based on the change in the feature. Detect points. Further, the polishing end point is detected by a logical operation combining a plurality of these feature amounts.

FIG. 4 shows two signal waveforms for explaining the characteristic amount. These signal waveforms are spectral waveforms, and the lower curve corresponds to signal number 33 in FIG. Here, a maximum value and a minimum value are selected as the feature amounts. On this signal waveform, the maximum value is indicated by ◇, and the minimum value is indicated by +. The maximum value and the minimum value can be calculated (extracted) by performing smoothing differentiation on the signal waveform. The positions (wavelength, reflectance) of the local maximum value and the local minimum value are feature amounts. In this embodiment, the reflectance is used. In extracting these features, the magnitude of the signal waveform is preferably normalized. This standardization is performed independently of changes in the polishing state of the wafer, such as fluctuations in the intensity of the irradiation light source, fluctuations in the transmittance of the optical system including lenses, fluctuations in the light-receiving sensitivity of the light-receiving unit, and fluctuations in the slurry. This is performed to reduce the influence of a fluctuating disturbance component on the signal. As a standardization method, a reference point is designated in a signal waveform, and the size of the signal waveform is corrected so that the size of the reference point is used as a reference value. As the reference point, when the signal waveform is a spectral waveform, the reflectance at a predetermined wavelength in a predetermined spectral range, the maximum value of the reflectance in the predetermined spectral range, or the maximum value in the predetermined spectral range. It is preferable to select one selected from the group of the reflectance, but the present invention is not limited thereto. In the example of FIG. 4, the normalization is performed such that the maximum value of the signal waveform is a predetermined reference value (in this case, 1). Specifically, waveform normalization can be performed by dividing the signal waveform by the largest local maximum value among a plurality of local maximum values. The upper curve in FIG. 4 shows the normalized signal waveform. This normalization of the signal waveform is preferably performed not only when extracting a maximum value or a minimum value as a feature value, but also when extracting any other feature value. It is preferable to perform the processing on the normalized waveform.

Further, when extracting the characteristic amount, it is preferable that after the normalization, the signal waveform is rotationally corrected about the standardized reference point. The reason for doing this is to remove the effects of slurry from the signal waveform. Since the reflected signal light is transmitted through the slurry, the acquired signal waveform contains a component which fluctuates under the influence of scattering or the like due to the slurry. This variation is proportional to the slurry concentration and is wavelength dependent. Since the fluctuation amount is generally larger on the shorter wavelength side, the higher the slurry concentration, the stronger the signal waveform tends to rise to the right. This is shown in FIG. In the state of FIG. 15 (b) and FIG. 15 (c) of this signal waveform,
When gma and SumPB are extracted, their values are different from those in the case of FIG. 15A when there is no slurry, and are dependent on the slurry concentration. In other words, the magnitude of the feature amount is affected by the slurry concentration other than the information inherent in the wafer such as the film thickness, and the measurement accuracy at the process end point is reduced. Therefore, the signal waveform is corrected, and the signal waveform is returned to the state shown in FIG. As a correction method, the signal is rotated around a standardized reference point (point indicated by each X mark in the upper right) of the signal waveform in FIG. 15B or FIG. At this time, the slope is determined by approximating the signal waveform by a linear curve and determining the slope. As a method of correcting the rotation of the signal waveform, there is a method of dividing the signal waveform by the characteristic of the slurry, using the characteristic of the slurry separately measured by a blank mirror or the like as a reference value, in addition to the method by this rotation. Of course, in the case of this second method, normalization is performed after this division.

As another characteristic amount, adjacent features in the signal waveform are used .
| Maximum value-minimal value |
Is | maximum-maximum for each of the plurality of maximum value / minimum value pairs.
Addition of small value |, that is, Σ | maximum value−minimum value |
The integral value of the waveform, or the first derivative of each feature,
Alternatively, the second derivative of each of the feature amounts can be used.

Here, in particular, the feature amount obtained by adding each of the | maximum value−minimum value | to the plurality of the maximum value / minimum value pairs is a difference between the maximum value and the minimum value (Sum OF Pe).
akato Bottom)
Called B. In FIG. 4, SumPB is the sum of elevation differences between peaks and valleys corresponding to adjacent ◇ and + of the normalized waveform, and is ((（1) − (+ 1)) + ((◇ 2) − (+2) +
((◇ 3) − (+ 3)).

Further, the integral value of the signal waveform is simply represented by S
igma. In FIG. 4, Sigma is an area surrounded by the normalized waveform, the wavelength axis, and the vertical axis (reflectance axis). Further, the time derivative of SumPB or Sigma
When the time derivative of “a” is used as a feature value, and this is applied to the case of FIG. 3, the time derivative of SumPB is calculated for the normalized signal (only the original signal is shown in FIG. 3) at each signal number. The slope of SumPB, that is, the difference of SumPB between adjacent signal numbers (for example, between 44 and 45). The time derivative of Sigma is the gradient of Sigma with respect to the normalized signal at each signal number, that is, the difference of Sigma between adjacent signal numbers (for example, between 44 and 45).

By the way, the reflected light from the pattern surface of the semiconductor device wafer reflects the device (
It can be considered as a superposition of light waves from each layer and each part of the (stacked thin film), and the spectral waveform of the reflected signal light resulting from this superposition is due to complicated interference effects (for example, when the film thickness of the top layer is the same). (Even though) it is very different from the blank film. FIG. 13 is an explanatory diagram illustrating the concept of this interference. FIG. 13 shows a cross section of one device wafer. In FIG. 13, 18 is a metal electrode layer, 19 is a dielectric layer,
21 is a lower layer part, 20 is an irradiation light spot, and 10
Reference numerals 0, 200, 300, a, and b denote reflected light waves from each layer and each portion of the (laminated thin film) of the device, respectively. Since the result of the interference of these light waves with each other becomes a reflected signal light, is there.

It is generally not easy to directly calculate the thickness of the object to be measured from the signal waveform obtained from the reflected signal light and determine the polishing state. Further, in addition to the difficulty in analyzing the spectral waveform, there is a problem of disturbance that gives instability to the spectral waveform.

A major first of these disturbances is the slurry. In the case of FIG. 1, the slurry is the slurry adhered to the upper surface of the window plate 15 in the light transmitting window 5. The slurry layer thickness through which the irradiation light and the reflected signal light are transmitted fluctuates irregularly during polishing and the slurry component fluctuates irregularly.
Gives unpredictable noise to the signal waveform.

Secondly, as can be seen from FIG. 1, the rotation of the surface plate 4 causes the light-transmitting window to cross the irradiation light, and every time measurement is performed, the irradiation light spot moves to a position different from the previous irradiation position. Disturbance caused by irradiation and measurement. This disturbance gives unpredictable noise due to inevitable non-uniformity of the remaining film thickness on the wafer and measurement of different pattern types at different positions. [Embodiment 1] As described above, what is handled here is a signal that is difficult to analyze and is affected by disturbance.
For this purpose, the present invention has attempted to extract a plurality of feature amounts capable of capturing a polishing state from a signal waveform and logically operate these feature amounts using fuzzy inference.

In the fuzzy inference of the present embodiment, the following feature values are used. However, other feature value groups can also be used. These feature values are used for experiments or theoretical studies according to the type of wafer. Is selected based on

In this embodiment, SumPB,
Sigma, 1st derivative of SumPB, Sig
ma first derivative, SumPB second derivative,
These are six feature quantities of Sigma's second derivative.

The fuzzy rule is not particularly limited, and is appropriately selected based on experiments or theoretical studies according to the type of wafer and the like. In the present embodiment, the following two rules are used. These rules are linked by "again" in the fuzzy rules. Rule 1: If large, small, small, small, small, negative, and positive, the end points are close. Rule 2: is small, large, large, large, large, or positive,
If it is negative, the end point is far.

Here, each of "large" and "small" in rules 1 and 2 is based on a membership function. The upper figure of FIG. 6 shows SumPB, and the lower figure shows Sig.
FIG. 7 shows a membership function expression of the fuzzy rule.

Here, the membership function is a function indicating the degree of matching (the degree of matching) of the vague words “large” and “small” to the fact that it is large or small, in the fuzzy rule. The fuzzy inference here is based on the Sugano method (Sugeno, M., Industrial app.
lications of fuzzy control, Elsevier Science Pub.C
o., 1985). This membership function is determined in advance for each feature amount based on preliminary experiments, calculation results, and the like.

In the upper part of FIG. 6, the horizontal axis represents the value of SumPB, and the vertical axis represents the degree of matching (degree of matching). Su
When the value of mPB is 1.6 or more, the membership function of “large” is 1 and the membership function of “small” is 0, which means that when the value of SumPB is 1.6 or more, The matching degree between the value of SumPB and “large” is 1, and the matching degree with “small” is 0. When the value of SumPB is 0.8 or less, “small”
Is 1 and the membership function of “large” is 0, which means that the value of SumPB is 0.
When the value is 8 or less, the degree of matching of the value of SumPB with “small” is 1, and the degree of matching with the value of “large” is 0. Further, when the value of SumPB is more than 0.8 and less than 1.6, both the degree of matching with “large” and the degree of matching with “small” take a value of 0 or more and 1 or less. However, it is an area that cannot be said to be “large” or “small”.

In the lower diagram of FIG. 6, Sigma rule 1
The membership functions of “small” and “large” in Rule 2 are shown, and their meanings may be interpreted in the same manner as in SumPB.

Next, in FIG. 7, (1), (2), (3),
(4), (5) and (6) correspond to the SumPB, Si
gma, the first derivative of SumPB (SumPB-Di
ff), the first derivative of Sigma (Sigma-Di)
ff), the second derivative of SumPB (SumPB-Di)
ff2), and the second derivative of Sigma (Sigma)
-Diff2). Of the upper and lower two columns, the upper column is for rule 1 and the lower column is for rule 2. Where (1) and
(2) shows the membership function shown in FIG. 6 divided into rule 1 and rule 2 and reduced and displayed.
In each of the membership functions of (3), (4), (5), and (6), the horizontal axis represents the value of each feature, and the vertical axis represents the degree of matching with each of rule 1 and rule 2 ( 0-1). Also,
Each straight line parallel to each vertical axis of each membership function is an input value of each feature amount of,,, and for a certain signal number, and is 2.50, 75, 0.12,
2.50, -1.00, and 1.00. Each intersection between each of these straight lines and each of the membership functions is the degree of matching of each feature.

(1), (2), (3), (4) for rule 1
, (5), and (6) are 1, 1, 0.6, respectively.
0, 0.75, 1, and 1, and Rule 1 takes the logical product of these. In the present embodiment, an algebraic product is taken as a logical product, and the result of Rule 1 is 1 × 1 × 0.60 × 0.7
5 × 1 × 1 = 0.45. The result of this rule 1 is
In FIG. 7A, the degree of matching with respect to 1 when the polishing end point is set to 1 is shown as 0.45, which is a membership function resulting from Rule 1.

Further, (1), (2), (3) for rule 2
, (4), (5), and (6) are 0, 0,
0.4, 0.25, 0, and 0, and Rule 2 takes the logical sum of these. Algebraic sum is used as the logical sum, and the result of rule 2 is 0 + 0 + 0.4 + 0.25 + 0 + 0−
(0.4 × 0.25) = 0.55. This rule 2
7B shows the degree of matching with respect to 0 when the completely unfinished point of polishing is set to 0 in FIG. 7B, which is 0.55, which is a membership function of the result of Rule 2. is there.

Next, if the result of rule 1 and the result of rule 2 are expressed together in FIG. 7 (c), the result of rule 1 and the result of rule 2 are linked by "again". This is the end result of the fuzzy inference, again a membership function.

Next, it is preferable to perform defuzzification to extract the essence from the membership function shown in FIG. 7C. This defuzzification method preferably obtains the center of gravity of the membership function of the final result, but is not limited to this method. When obtaining from the center of gravity, the following equation is used to calculate the center of gravity = 1 × 0.45 + 0 × 0.55 / 0.45 + 0.5
5 = 0.45 0.45, which is the value at the time of polishing (signal number)
Is used as the end point evaluation value in

In this fuzzy inference, the closeness of the polishing end point,
That is, the end point evaluation value is indicated by a value in the range from 0 to 1,
It is known in advance that the time when the value reaches 0.9 or more is the polishing end point. FIG. 8 shows how the end point evaluation value changes.
In FIG. 8, the horizontal axis represents the signal number, and the vertical axis represents the end point evaluation value (0
Since the end point evaluation value is 0.9 or more when the signal number is 33, this can be determined as the polishing end point, and the polishing end point signal can be output.

Next, a detailed description will be given of a change in a feature value extracted from a signal waveform, which is a basis for performing the fuzzy inference. FIG. 9 shows an example of the change of SumPB. The horizontal axis indicates the signal number, which corresponds to the number of times of polishing. The dotted line indicates the value of SumPB, and the solid line indicates the value of the moving average of SumPB. FIG. 10 shows an example of the change in Sigma. The meanings of the dotted line, solid line, and horizontal axis are the same as in FIG. Su
For both mPB and Sigma, the input to the fuzzy rule uses the value of the moving average.

Rules 1 and 2 for fuzzy inference described above
Among them, the rule of “large” and “small” in SumPB is a rule for determining the magnitude of the change in the signal waveform, and the rule of “large” and “small” in Sigma is the rule of the entire signal waveform. Since this is a rule for determining the degree of the size, this part can be said to be a quantitative rule for feature amount determination. In addition, 1 of SumPB
Derivative, Sigma's first derivative, SumP
The second derivative of B and the second derivative of Sigma are rules for finding local maxima and minima in the curves of SumPB and Sigma in FIGS. 9 and 10, and a qualitative rule for grasping the shape of the curves. I can say.

Regarding the quantitative part and the value, the value greatly varies depending on the type of wafer and the type and state of slurry, and therefore the value of SumPB and Sigma with the progress of polishing are large.
It is preferable to perform tuning to shift the membership function sideways during measurement according to the change in the value of. As a reference value for tuning, it is preferable to use an average value of the feature amounts from the start of polishing to the time of measurement. Each solid line parallel to the horizontal axis in the graphs of FIGS. 9 and 10 indicates the average value. Thus, for example, at each measurement stage during polishing, Sum
After calculating the average value of PB, Sum
Tuning is performed, for example, by shifting the membership function shown in the upper part of FIG. 6 horizontally so that both the degree of matching with the PB membership function and the degree of matching with the “large” and “small” become 0.5. Tuning of the Sigma membership function is also performed by shifting the membership function shown in the lower diagram of FIG. 6 horizontally, as in the case of SumPB.

By this tuning, SumPB and S
Even when the value of igma changes due to fluctuation of the slurry, etc., the membership function can be appropriately selected.
As described above, in the present invention, two or more features are extracted from a signal waveform, and a logic operation is performed on these features using fuzzy inference, so that the measurement target is a substrate having a device pattern. Even if there is disturbance due to fluctuations in the slurry or the measurement position, the polishing end point can be detected with high accuracy and simultaneously with polishing.

When the signal processing operation of the signal processing device in the above description is performed by a computer, the operation is shown in FIG. Hereinafter, the operation of the signal processing device will be described with reference to step numbers.

First, when the signal processing device is turned on, the CPU 31 in FIG. 5 acquires an optical signal (S1). This optical signal is obtained at each sampling period interval. Next, C
The PU 31 extracts a feature amount from the optical signal (S2). Prior to the extraction of the feature, the feature is selected (S10). This selection may be made manually or automatically selected in accordance with the type of wafer or the like.

Next, the CPU 31 tunes the membership function (S3). Prior to this S3,
A membership function has been determined (S11). This determination may be made manually or automatically selected in accordance with the type of wafer or the like.

Next, the CPU 31 calculates each degree of coincidence with respect to the input value of each feature amount (S4). Next, CPU3
1 calculates the result of the fuzzy rule (S5).

Prior to the step of S5, a fuzzy rule has been determined (S12). This decision method is
Depending on the wafer type or the like, it may be performed manually or determined automatically.

Next, the CPU 31 calculates the final result of the fuzzy inference by combining the results of the respective fuzzy rules (S6). Next, the CPU 31 defuzzifies the final result of the fuzzy inference (S7).

Next, the CPU 31 determines whether or not the value of the defuzzification has reached a value preset as a process end point (S8). Prior to S8, the value of the process end point is set (S13). This setting may be performed manually or determined automatically according to the type of wafer.

If NO in S8, the CPU 31 performs a process on an optical signal sampled and acquired next. If YES in S8, CPU 31 outputs a process end point signal (S9).

In the present embodiment, the end point of polishing is detected by fuzzy inference using the extracted feature values.
Even if the signal is disturbed or the wafer has a pattern formed thereon, one or both of high-precision and stable detection of the process end point and / or simultaneous detection can be performed. [Embodiment 2] In the measurement of Embodiment 1, when the logical operation of two or more feature values is performed, the polishing end point is detected by using fuzzy inference. However, when the disturbance of the signal waveform is small, etc. If the required accuracy can be obtained without using fuzzy inference, or if fuzzy inference is used, the logical operation becomes complicated and the cost may increase. In these cases, no fuzzy inference is used. For example, instead of fuzzy rule 1 in the fuzzy inference described above, Su
mPB is greater than threshold value S1 and Sigma is
2 and the first derivative of SumPB is smaller than the threshold S3, the first derivative of Sigma is smaller than the threshold S4, and the second derivative of SumPB is a negative value, and An equation of a logical operation can be an equation in which the polishing end point is set when each of these feature values satisfies all of the above-mentioned conditions, such as a polishing end point if the second derivative is a positive value. Here, S1, S2, S3, and S4 are constant values determined for each wafer.

When the signal processing operation of the signal processing apparatus in the above description is performed by a computer, the operation is shown in FIG. Hereinafter, the operation of the signal processing device will be described with reference to step numbers.

First, when the signal processing device is turned on, the CPU
31 acquires an optical signal (S31). This optical signal is
It is acquired at each sampling cycle interval. Next, CPU
31 extracts a feature amount from the optical signal (S32). Prior to the extraction of the feature, the feature is selected (S
36). This selection may be made manually or automatically selected in accordance with the type of wafer or the like.

Next, the CPU 31 performs a logical operation (S
33). Prior to S33, a logical operation algorithm is determined (S37). This determination may be made manually or automatically selected in accordance with the type of wafer or the like.

Next, the CPU 31 determines whether or not the result of the logical operation satisfies the process end point condition (S34).
If NO in S34, the CPU 31 performs a process on the optical signal sampled and acquired next.

If YES in S34, CPU 31 outputs a process end point signal (S35). In this embodiment,
Since the polishing end point is detected by a logical operation using the extracted feature quantity, even if there is a disturbance in the signal or the wafer on which the pattern is formed, it does not reach the case of the first embodiment, One or both of high-precision and stable detection of the process end point, or simultaneous detection can be performed. [Embodiment 3] In the measurements of Embodiments 1 and 2 described above, two or more feature amounts are selected from the signal waveform, and the polishing end point is detected based on these logical operations. Depending on the type), there is a case where a logical operation is rather unfavorable, or a case where performing a logical operation becomes a problem in terms of cost. In this case, only one feature value is selected, and the change is detected. The selected feature amount is | maximum value−minimum value | for a pair of adjacent maximum value / minimum value in a signal waveform (in this case, a spectral waveform), or each | maximum value of a plurality of the maximum value / minimum value pairs. Either the sum of the minimum values | or the integral value of the signal waveform is preferable. In this case, the measurement is simplified in polishing the wafer having the pattern.

When the signal processing operation of the signal processing device in the above description is performed by a computer, the operation is shown in FIG. Hereinafter, the operation of the signal processing device will be described with reference to step numbers.

First, when the signal processing device is turned on, the CPU
31 acquires an optical signal (S41). This optical signal is
It is acquired at each sampling cycle interval. Next, CPU
31 extracts a feature amount from the optical signal (S42). Prior to the extraction of the feature, the feature is selected (S
45). This selection may be made manually or automatically selected in accordance with the type of wafer or the like.

Next, the CPU 31 determines whether or not the feature amount has reached a set value (S43). Prior to S43, the value of the process end point is set (S46). This setting may be performed manually or automatically according to the type of the wafer.

If NO in S43, the CPU 31 performs a process on an optical signal sampled and acquired next. If YES in S43, CPU 31 outputs a process end point signal (S44).

FIG. 19 shows an example measured by the present measuring method. FIG. 19 shows a TEG (TestElement Gr).
oove) Sigma for the signal number when Sigma and SumPB are selected as the feature amounts for the pattern
And changes in SumPB. In the case of the present embodiment, the point of time of signal number 50 corresponds to the polishing end point, and at this point, Si
Since both gma and SumPB change rapidly, the polishing end point can be detected by capturing this timing. In this case, the feature amount (in this case, S
igma and SumPB) by performing the differentiation once or twice, so that the detection of the polishing end point is further facilitated.
It is preferable to use one time and two times differentiation of a signal together.

In this embodiment, since the polishing end point is detected based on the change in the extracted feature value, it is not necessary to use a logical operation algorithm or fuzzy inference. Even in this case, one or both of the end points of the process can be simply and accurately detected, or the simultaneous detection can be performed. Further, depending on the type of device pattern, detection can be performed with higher accuracy than in the case of performing a logical operation.

The measuring apparatus using the measuring method described in the first, second, and third embodiments is provided in a polishing apparatus or the like, and is used for measuring a process state. [Embodiment 4] This embodiment relates to a method of manufacturing a semiconductor device using the polishing apparatus of the present invention.

FIG. 11 is a flowchart showing a semiconductor device manufacturing process. After the semiconductor device manufacturing process is started, first, in step S200, an appropriate processing step is selected from the following steps S201 to S204. Steps S201 to S20 according to the selection
Go to any of 4

Step S201 is an oxidation step for oxidizing the surface of the silicon wafer. Step S202 is CV
D to form an insulating film on the silicon wafer surface
This is a VD process. Step S203 is an electrode film forming step of forming an electrode film on a silicon wafer by a process such as vapor deposition. Step S204 is an ion implantation step of implanting ions into the silicon wafer.

After the CVD step or the electrode film forming step, the flow advances to step S209 to determine whether or not to perform the CMP step. If not, the process proceeds to step S206; otherwise, the process proceeds to step S205. Step S2
Reference numeral 05 denotes a CMP step, in which a polishing apparatus of the present invention is used to planarize an interlayer insulating film and to form a damascene by polishing a metal film on the surface of a semiconductor device.

After the CMP step or the oxidation step, the process proceeds to step S206. Step S206 is a photolithography step. In the photolithography process, a resist is applied to a silicon wafer, a circuit pattern is printed on the silicon wafer by exposure using an exposure device, and the exposed silicon wafer is developed. Further, in the next step S207, portions other than the developed resist image are removed by etching.
Thereafter, the resist is stripped, and the etching step is performed to remove the unnecessary resist after the etching.

Next, it is determined in step S208 whether all necessary processes have been completed, and if not, step S20
Returning to 0, the above steps are repeated to form a circuit pattern on the silicon wafer. If it is determined in step S208 that all steps have been completed, the process ends.

In the semiconductor device manufacturing method according to the present invention, since the polishing apparatus according to the present invention is used in the CMP step, the accuracy of detecting the polishing end point in the CMP step is improved, and the yield in the CMP step is reduced. improves.
As a result, there is an effect that a semiconductor device can be manufactured at a lower cost than a conventional semiconductor device manufacturing method.

The present invention can be used in a CMP step of a semiconductor device manufacturing process other than the semiconductor device manufacturing process shown in FIG. The semiconductor device according to the present invention is manufactured by the semiconductor device manufacturing method according to the present invention. As a result, a semiconductor device can be manufactured with higher quality and at lower cost than the conventional semiconductor device manufacturing method, and the manufacturing cost of the semiconductor device can be reduced.

The invention of the first, second, third, and fourth embodiments has been described above. However, a function that enables any two or more measurement methods selected from the signal processing methods of the first, second, and third embodiments. May be incorporated in one measuring device, and any one of the functions may be selected and used for the measurement. This makes it possible to select an optimal measurement method for the type of wafer and polishing conditions.

The present invention is not limited to the case where the measurement is performed through the light-transmitting window as shown in FIG. 1, but also enables the polishing head to swing in addition to the rotation so that the wafer protrudes from the polishing pad, and the protruding portion is exposed to light. Irradiation and measurement. In this case, no translucent window is required. Further, in a polishing apparatus in which the polishing pad is smaller than the wafer, the measurement can also be performed on a portion where the wafer protrudes from the polishing pad and is exposed.

The present invention is not limited to the polishing end point,
It can also be used for detecting the end point of other removal processes such as ion etching and the like, and the film formation process such as CVD and sputtering. Further, the terminating point of the process referred to here includes not only the terminating point of the step in a general thin film removing step but also the terminating point of an intermediate step such as the timing at which the removing step proceeds to a different material layer.

Although the measuring apparatus shown in FIG. 1 emits light from the pattern surface side of the semiconductor device, the light can be emitted from the back side of the wafer. In this case, the light source requires a multi-wavelength component light source in the infrared region.

Although the present invention has been described with reference to the drawings, the scope of the present invention is not limited to the ranges shown in these drawings, and the present invention is limited to the above description. Not even a thing.

[0098]

According to the present invention, the signal waveform is
Extract features that change appropriately in response to changes in thin film thickness
And perform logical operation, so even if there is a pattern on the board,
Polished layer clearly changes even if there is some disturbance in signal waveform
High accuracy detection of process end point or simultaneous
One or both of the detections can be performed. [Claim 2] In the present invention, since fuzzy inference is used for a logical operation, even if there is a large disturbance in a signal waveform, one or both of high-precision detection of a process end point and simultaneous detection can be performed. [Claim 3] In the present invention, the membership function is tuned during the measurement, so that a membership function suitable for the magnitude of the extracted feature can be set at any time. One or both of high-precision detection of points and / or simultaneous detection can be performed. [Claim 4] In the present invention, a process end point is measured by monitoring a change in a feature amount that reflects a process state extremely well, and a logical operation is not performed, so that the process end point is detected simply and with high accuracy. One or both of the simultaneous detections can be performed. Further, it is most suitable depending on the type of the device pattern. [Claim 5] In the present invention, according to any one of claims 1 to 4,
In addition to the bright effect, the feature amount is extracted from the normalized waveform, so that even if there is disturbance, one or both of the process end points can be detected simply and with high accuracy, or simultaneous detection. [Claim 6] In the present invention, even if the signal waveform is tilted due to an abrasive or the like, the feature amount is extracted from the waveform obtained by rotation-correcting the signal waveform. You can do one or both. [Claim 7] In the present invention, a change in the thickness of a thin film from a signal waveform.
The feature values that change appropriately in response to the above are extracted and logically operated, so even if there is a pattern on the substrate, there is some disturbance in the signal waveform, or the polishing layer does not change clearly, Either one or both of high-precision detection and / or simultaneous detection of the end point can be performed. [Claim 8] In the present invention, a process end point is measured by monitoring a change in a feature amount that reflects a process state extremely well, and a logical operation is not performed, so that the process end point can be detected simply and with high accuracy. One or both of the simultaneous detections can be performed. Further, it is most suitable depending on the type of the device pattern. [Claim 9] In the present invention, since the measuring apparatus according to any one of claims 1 to 6 is provided, a wafer having a device pattern can be polished with high precision or with good yield. [Claim 10] In the present invention, a semiconductor device is polished by the polishing apparatus according to claim 9, so that a semiconductor device can be manufactured with high quality or at low cost. [Claim 11] According to the present invention, it is possible to realize, on a computer, the "feature amount extraction unit and logical operation unit according to any one of claims 1 to 6 ".

[Brief description of the drawings]

FIG. 1 is a schematic diagram of a CMP polishing apparatus of the present invention.

FIG. 2 shows an example of a signal waveform (spectral waveform).

FIG. 3 is an example of continuous display of a signal waveform (spectral waveform).

FIG. 4 is a diagram showing a relationship between a signal waveform, a normalized signal waveform, and a feature amount;

FIG. 5 is a diagram of a signal processing unit using a computer.

FIG. 6 shows examples of membership functions of SumPB and Sigma.

FIG. 7 is an example of a membership function expression of a fuzzy rule.

FIG. 8 is a graph of end point evaluation values with respect to signal numbers.

FIG. 9 is a graph of the signal number of SumPB.

FIG. 10 is a graph of Sigma signal numbers.

FIG. 11 is a flowchart illustrating an example of a semiconductor device manufacturing process.

FIG. 12 is a schematic view of a conventional CMP polishing apparatus.

FIG. 13 shows the relationship between the reflected light wave from each layer and each part of the device and the irradiation light spot.

FIG. 14 is an explanatory diagram of a minimum unit of a pattern.

FIG. 15 is a diagram showing an outline of rotation of a signal waveform.

FIG. 16 is a flowchart of an embodiment of signal processing based on fuzzy inference.

FIG. 17 is a flowchart of an embodiment of signal processing by a logical operation.

FIG. 18 is a flowchart of an embodiment of signal processing based on a change in a feature value.

FIG. 19: Changes in Sigma and SumPB

[Description of Signs] 1 Polishing head 2 Substrate (wafer) 3 Polishing pad 4 Surface plate 5 Transparent window 6 Light receiving unit 7 Irradiation light and reflected light 8 Signal processing unit (personal computer) 9 White light source 10 Beam splitter 11 to 13 Lens Reference Signs List 15 transparent window material 16 abrasive (slurry) supply mechanism 17 abrasive (slurry) 18 metal electrode layer 19 dielectric layer 20 irradiation light spot 21 lower layer part 30 polishing end point measuring device a, b, 100, 200, 300 reflected light wave ωH Indicates rotation of polishing head ωT Indicates rotation of platen

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat ゛ (Reference) // G01B 11/06 H01L 21/306 U F term (Reference) 2F065 BB03 CC19 DD04 FF46 GG02 GG03 GG07 GG23 GG24 HH04 LL46 LL67 QQ13 QQ17 QQ27 QQ29 QQ32 QQ34 QQ42 3C058 AA09 AC02 AC04 BA01 BB09 CB03 CB05 DA12 DA17 4M106 AA01 BA04 BA07 BA08 CA38 CA70 DB03 DB07 DB12 DH03 DH12 DH31 DH38 DJ12 DJ13 DJ17 5F3 DD

Claims (12)

[Claims] An end point of a step of forming an insulating film or a metal electrode film on a substrate or a step of removing the film is defined as:
A measurement device that irradiates the substrate surface with light and measures a signal waveform obtained by detecting one or both of the reflected signal light and the transmitted signal light, and extracts two or more feature amounts from the signal waveform. A measurement apparatus comprising: a feature amount extraction unit; and a logic operation unit that performs a logic operation using the two or more feature amounts to determine a process end point. 2. The method according to claim 1, wherein the signal waveform is a spectral waveform, and the characteristic amount is a maximum value, a maximum maximum value, a minimum value, a minimum minimum value, a maximum value / minimum value in the signal waveform. , The maximum / minimum value, the | maximum value-minimum value | of the adjacent maximum value / minimum value pair, and the added value of each | maximum value-the minimum value | An integrated value of the signal waveform, a group of one and two time differential coefficients of each of the feature amounts, a group of positive and negative signs of the time differential coefficient,
The measuring device according to claim 1, wherein the measuring device is selected from a feature amount group consisting of: 3. The measuring device according to claim 1, wherein the logical operation unit makes the determination using fuzzy inference. 4. A measuring apparatus according to claim 3, wherein a membership function used in said fuzzy inference is tuned during measurement by a value calculated from said characteristic amount. 5. A process for forming an insulating film or a metal electrode film on a substrate, or a process end point in a process for removing the film,
A measuring device that irradiates the substrate surface with light, and measures from a change in a feature amount extracted from a signal waveform obtained by detecting one or both of the reflected signal light and the transmitted signal light,
A feature value extracting unit for extracting a feature value from the signal waveform, wherein the signal waveform is a spectral waveform, and the feature value is | maximum with respect to an adjacent maximum value / minimum value pair in the signal waveform. Value-minimum value |, or an addition value of each | maximum value-minimum value | with respect to a plurality of said maximum value / minimum value pairs, or an integrated value of the signal waveform. 6. The method according to claim 1, wherein the characteristic amount is extracted from a waveform obtained by normalizing the signal waveform.
The measuring device according to the item. 7. The measuring apparatus according to claim 1, wherein the characteristic amount is extracted from a waveform obtained by rotating and correcting the signal waveform. 8. A process for forming an insulating film or a metal electrode film on a substrate, or a process end point in a process for removing the film,
A method of irradiating the substrate surface with light and measuring from a signal waveform obtained by detecting one or both of the reflected signal light and the transmitted signal light, and extracting two or more features from the signal waveform. And a step of performing a determination by performing a logical operation using the two or more characteristic amounts. 9. A process for forming an insulating film or a metal electrode film on a substrate, or a process end point in a process for removing the film,
A measurement method of irradiating the substrate surface with light, and measuring one or both of the reflected signal light and the transmitted signal light by measuring a change in a feature amount extracted from a signal waveform obtained by detecting the signal waveform. , A spectral waveform, and the feature amount is | maximum value−minimum value || for a pair of adjacent maximum value / minimum value in the signal waveform, or | maximum value− of each of a plurality of the maximum value / minimum value pairs. A measurement value, which is an added value of the minimum value | or an integrated value of the signal waveform. 10. A polishing apparatus, comprising: a holder for holding a substrate; a polishing body; and the measuring device according to claim 1, wherein an abrasive is interposed between said substrate and said polishing body. In the state
A polishing apparatus characterized in that a process end point can be measured when a substrate is polished by applying a load between the substrate and the polishing body to give a relative motion between the two. . 11. A method for manufacturing a semiconductor device, comprising a step of polishing a surface of a semiconductor wafer using the polishing apparatus according to claim 10. 12. A machine readable recording machine storing a signal processing program for causing a computer to function as a "feature amount extraction unit and a logical operation unit or a feature amount extraction unit according to any one of claims 1 to 7". recoding media.